Profile Units

Time - ms

Mem - MB

Size - MB

Obj-Detection-Optimizer

1. Basic Optimizer Block

from optimizer import Optimizer
Optimizer(
    dep_filename="dep.csv",
    prof_filenames=[
        "prof.csv",
        "prof.csv",
        "prof.csv",
        "prof.csv",
        "prof.csv",
    ],
    bandwidth=2000,
    parallel=True,
    ignore_latency=True,
    iterations=1,
    dir="testcase/explore"
)

2. Attribute Explanation

1. dep_filename

• This csv file contains the dependency relation between layers of a network.
• It has two columns: source, destination. Every entry represents an edge in the network.

2. prof_filenames

• This csv file contains the profiling result of every layer on a particular device.
• One entry in this list represents one device. For example, in the basic optimizer block above, there are five devices available.
• It has five columns: layer_name, time,cpu_mem, cuda_mem, size, MACs

3. bandwidth

• The bandwidth of communication network between drones. The unit is MBps.

4. parallel

• data-computation parallel.

5. ignore_latency

• Whether to ignore transfer latency. Mainly for testing.

3. Use Optimizer Wrapper

Please use opt_wrapper.py when optimizing the network. Make a copy of the prof and dep files to the root directory.

You may modify the following part as needed.

bandwidth = 400
ignore_latency = False
iteration = 5
prof_filenames = [
        "prof.csv",
        "prof.csv",
        # "prof.csv",
        # "prof.csv",
        # "prof.csv",
        # "prof.csv",
        # "prof.csv",
        # "prof.csv",
    ]
benchmark = 114.46748520400001  # agx
# ...